阅读说明：本技术 一种三自由度混合磁轴承转子位移自检测方法 (Three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method ) 是由 刘钙 吴亦然 于 2020-07-21 设计创作，主要内容包括：本发明公开一种三自由度混合磁轴承转子位移自检测方法,属于高速及超高速电机传动领域,本发明是利用混合核函数支持向量机位移预测模型实现三自由度交直流混合磁轴承位移自检测的方法,以磁轴承控制电流为输入样本,径向和轴向位移为输出样本,采集样本数据,选用混合核函数,通过粒子群算法优化支持向量机性能参数,利用训练样本和性能参数训练最小二乘支持向量机,建立非线性预测模型,将预测模型串接到三自由度交直流混合磁轴承之前,与线性闭环控制器相连,加上扩展的电流滞环三相功率逆变器和开关功率放大器共同构成磁轴承位移闭环控制,实现三自由度交流混合磁轴承无位移传感器自检测。(The invention discloses a self-detection method for rotor displacement of a three-degree-of-freedom hybrid magnetic bearing, which belongs to the field of high-speed and ultra-high-speed motor transmission and is a method for realizing self-detection of the displacement of a three-degree-of-freedom alternating current-direct current hybrid magnetic bearing by utilizing a hybrid kernel function support vector machine displacement prediction model, using magnetic bearing control current as input sample, radial and axial displacement as output sample, collecting sample data, selecting mixed kernel function, optimizing performance parameters of a support vector machine by a particle swarm algorithm, training a least square support vector machine by utilizing a training sample and the performance parameters, establishing a nonlinear prediction model, connecting the prediction model in series before the three-degree-of-freedom alternating current-direct current hybrid magnetic bearing, the three-phase power inverter is connected with the linear closed-loop controller, and the extended current hysteresis three-phase power inverter and the switching power amplifier jointly form magnetic bearing displacement closed-loop control, so that the self-detection of the three-degree-of-freedom alternating-current hybrid magnetic bearing displacement-free sensor is realized.)

1. A three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method is characterized by comprising the following steps:

step one, taking the control current of the magnetic bearing as an input sample, taking radial and axial displacement as output samples, and collecting sample data;

selecting a mixed kernel function, optimizing performance parameters of a support vector machine through a particle swarm algorithm, training a least square support vector machine by using a training sample and the performance parameters, and establishing a nonlinear prediction model;

and step three, connecting the prediction model string with a linear closed-loop controller, and forming magnetic bearing displacement closed-loop control by combining an expanded current hysteresis three-phase power inverter and a switching power amplifier, so as to realize the self-detection of the three-degree-of-freedom alternating-current hybrid magnetic bearing displacement-free sensor.

2. The three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method according to claim 1, wherein the first step is specifically:

continuous acquisition of q sets of current signals x in a magnetic bearing system with eddy current sensors_t＝[i_at,i_bt,i_ct,i_zt]，

(t ═ 1,2,3.., q) and rotor displacement signal y_t＝[x_ft,y_ft,z_ft](t ═ 1,2,3.., q) as initial input and output sample data, where i is output as initial input_atIs the current signal of the magnetic pole A, i_btIs the current signal of the magnetic pole B, i_ctIs the current signal of the magnetic pole C, x_ftIs a displacement signal in the radial x-direction, y_ftIs a displacement signal in the radial y-direction, z_ftIs the axial z-direction displacement signal); selecting q/2 groups as training sample set, current signal x_t＝[i_at,i_bt,i_ct,i_zt](t 1,2,3.., q/2) and a rotor displacement signal y_t＝[x_ft,y_ft,z_ft](t ═ 1,2,3.., q/2), for offline training samples; the other half q/2 group is used as a test sample set for testing the sample set.

3. The three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method according to claim 1, wherein the second step specifically comprises:

2.1, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt]Inputting the parameters into a particle swarm algorithm and a support vector machine respectively to determine parameters of a mixed kernel function in the support vector machine;

2.2, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt]Inputting the (t ═ 1,2,3.., q/2) into a particle swarm algorithm, optimizing a mixing coefficient lambda by adopting the particle swarm algorithm, reserving the lambda with the best performance index as an optimal performance parameter, and then inputting the lambda with the best performance index into a support vector machine;

2.3, q/2 sets of current signals x_t＝[i_at,i_bt,i_ct,i_zt]And (t 1,2,3, q/2) inputting into a support vector machine, and determining the support vector machineThe mixed kernel function of (1);

radial basis kernel function of K_l＝exp(-Px-x_tP²/2²) (t ═ 1,2,3.., q/2), where x is the input current signal, x is_tIs the input current signal of the training sample set, is the kernel width, by bringing the input current signal x into the test sample set_t＝[i_at,i_bt,i_ct,i_zt](t 1,2,3, q/2) obtaining a mapping characteristic map of the radial kernel function when the kernel widths respectively take different values, and constructing the radial kernel function K by using the one with the best reserved characteristic_l＝exp(-Px-x_tP²/2²)，

(t＝1,2,3...,q/2)；

Polynomial kernel function of K_g＝((x,x_t)+1)^e(t ═ 1,2,3.., q/2), where x is the input current signal, x is_tIs the input current signal of the training sample set, e is the exponent, by bringing the input current signal into the test sample set

x_t＝[i_at,i_bt,i_ct,i_zt](t ═ 1,2,3.., q/2.) can obtain a mapping characteristic map of the polynomial kernel function when the index e takes different values respectively, and the one e with the best retained characteristic constructs the polynomial kernel function K_g＝((x,x_t)+1)^e，(t＝1,2,3...,q/2)；

Forming a mixed kernel function K by the optimal radial basis kernel function and the optimal polynomial kernel function_m＝λK_l+(1-λ)K_g，

(t ═ 1,2,3., q/2), where λ is the mixing coefficient that is optimal for the performance index obtained;

2.4, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt](t ═ 1,2,3.., q/2) and rotor displacement signals

y_t＝[x_ft,y_ft,z_ft](t 1,2,3.., q/2) is substituted into the prediction output functionIn, obtainObtaining corresponding support vector coefficient a_tAnd (t is 1,2,3.., q/2) and a threshold b, and further establishing a support vector machine displacement prediction model reflecting the current-displacement relation of the magnetic bearing, so that the prediction output of the output displacement can be identified according to the current input x of the modelWherein K_mAs a mixed kernel function, K_m＝λexp(-Px-x_tP²/2²)+(1-λ)((x,x_t)+1)^e(t ═ 1,2,3.., q/2), λ is a mixing coefficient.

4. The three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method according to claim 3, wherein in the particle swarm optimization, each sample represents a particle, the population size is q/2, the q/2 th sample represents the q/2 th particle, and the coordinate position vector of each particle is expressed as a vector in a target search space in D dimensionThe velocity vector is expressed as

1) initializing a sample group and calculating a fitness function value;

2) in the implementation process of the particle swarm optimization, when a better current optimal solution which cannot be obtained by the particle swarm optimization is obtained later, a certain amount of constant disturbance is added, so that the inertia weight is suddenly increased in certain iteration, local search is skipped conveniently, global search is carried out, and local convergence is prevented; for the linear decreasing particle swarm algorithm for increasing the disturbance inertia weight, the particle swarm algorithm updating iterative calculation formula for introducing the inertia weight coefficient is as follows:

A_s∈{0,0.1}

wherein t 1,2, L, q/2 represents the number of the particles; s represents the s-dimension of the particle, s 1, 2.., D; d represents the number of iterations; c. C₁,c₂Taking a value between 0 and 2 as an acceleration constant; rand is a random real number of interval (0, 1); omega_maxIs the initial maximum inertial weight; omega_kIs the decreasing slope of the inertial weight coefficient; a. the_sFor the inertial weight perturbation constant, at a perturbation probability of 10%, A_s0.1, the rest is A_s＝0；

3) The dynamic self-adaptive change inertia weight coefficient is as follows: under the probability of 40%, multiplying the inertia weight obtained by linearly decreasing the linear weight of the particle swarm algorithm for increasing the disturbance by a coefficient in a fixed range, wherein the coefficient r is in the interval of 0.9-1.1, namely r belongs to [0.9,1.1], and the specific dynamically adaptive particle swarm algorithm for changing the inertia weight coefficient updates an iterative calculation formula as follows:

A_s∈{0,0.1}

4) random individuals were introduced to maintain particle population diversity: according to the particle swarm updating mode, in the iterative process of the particle swarm algorithm, all individuals are close to the optimal particles, so that the particles of the particle swarm are aggregated to lose diversity; under the probability of 30%, correspondingly replacing a certain random individual in the solution space with the particles obtained by the particle swarm algorithm;

5) and obtaining the current optimal position, updating the particle position, judging whether the iteration stopping condition is met, if not, updating the particle position, recalculating the fitness function value, and if the iteration stopping condition is met, obtaining the optimal solution and outputting the optimal solution.

5. The three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method according to claim 1, wherein the third step is specifically:

connecting a mixed kernel function support vector machine displacement prediction model with a corresponding linear closed-loop controller, and connecting a corresponding expanded current hysteresis three-phase power inverter and a switching power amplifier behind the linear closed-loop controller in series to jointly form displacement closed-loop control on the three-degree-of-freedom AC/DC mixed magnetic bearing, thereby realizing the self-detection of the three-degree-of-freedom AC/DC mixed magnetic bearing displacement-free sensor; current signal i of three-phase power inverter and switching power amplifier with extended current hysteresis by mixing kernel function support vector machine displacement prediction model_at，i_bt，i_ct，i_ztAs an input signal, a predicted rotor displacement x is output_ft，y_ft，z_ftAnd with a given reference position signal x_ft*，y_ft*，z_ftComparing, and outputting rotor suspension control force F by linear closed-loop controller_xt*，F_yt*，F_ztThree-phase power from star to extended current hysteresisThe inverter and the switching power amplifier finally realize the stable suspension of the three-degree-of-freedom alternating current and direct current hybrid magnetic bearing.

Technical Field

The invention belongs to the field of high-speed and ultra-high-speed motor transmission, in particular to a three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method, which has wide application prospects in the fields of aerospace, vacuum technology, mechanical industry, energy traffic and the like.

Background

Magnetic bearings use magnetic field forces to levitate the rotor in the air without contact, and the levitation position can be controlled by a control system. Compared with the traditional bearing, the magnetic bearing has the outstanding advantages of no friction and wear, no need of lubrication, high rotating speed, high precision, long service life and the like, particularly in a high-speed machine tool spindle system, the supporting mode of the spindle determines the cutting speed, the processing precision and the application range which can be achieved by the machine tool to a great extent, and the magnetic bearing is applied to the support of the high-speed machine tool spindle, thereby creating favorable conditions for improving the technical level of the high-speed machine tool spindle.

The existence of the displacement sensor in the magnetic bearing occupies space, increases system cost, reduces the dynamic performance of the system, is not suitable for high-speed and high-precision occasions, and due to the structural limitation, the sensor cannot be arranged in the middle of the magnetic bearing, so that the control equations of the system are mutually coupled, and the design of a controller is complex, therefore, the research of the magnetic bearing self-detection technology is very necessary. Currently, the common self-detection technologies mainly include a high-frequency signal injection method, a PWM carrier analysis method, a differential transformer method, a kalman filter method, and the like. However, these methods are all completed based on the mathematical model of the magnetic bearing, and due to the intrinsic nonlinearity and parameter instability of the alternating-current hybrid magnetic bearing, it is difficult to establish an accurate calculation model of the rotor displacement. In recent years, a Support Vector Machine (SVM) created by Vapnik et al expresses arbitrary nonlinear mapping capability by its characteristics, can obtain a current global optimal solution according to limited sample information, has a fast training speed, a fixed topological structure and a strong generalization capability, can better solve the problems of nonlinearity, high dimension, local minimum and the like, and brings a new possibility to accurate prediction of rotor displacement, especially for small sample situations.

The Chinese patent publication No. CN103631138A discloses a displacement detection method for a three-degree-of-freedom hybrid magnetic bearing mixed kernel function support vector machine, aiming at the characteristics that the intrinsic nonlinearity and parameter instability of the hybrid magnetic bearing are overcome, and an accurate calculation model of rotor displacement is difficult to establish, a mixed kernel function support vector machine is utilized to establish a nonlinear prediction model between alternating magnetic bearing displacement and current, so that the self-detection of a three-degree-of-freedom alternating current and direct current hybrid magnetic bearing displacement-free sensor can be realized, the defects of the existing several common magnetic bearing self-detection technical methods are overcome, but the problems of local optimization and slow convergence rate are still existed in the particle swarm algorithm used for the prediction model force of the support vector machine, and the optimal kernel function performance parameters cannot be obtained, and the obtained mixed magnetic bearing kernel function support vector machine displacement prediction model is not accurate enough.

Disclosure of Invention

In order to solve the problems of over-learning, easy falling into local optimization and low convergence speed of a particle swarm algorithm, the invention provides a three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method, which is used for optimizing the particle swarm algorithm in a hybrid kernel function support vector machine displacement prediction model.

The invention is realized by the following steps:

a three-degree-of-freedom hybrid magnetic bearing rotor displacement self-detection method is characterized by comprising the following steps:

step one, taking the control current of the magnetic bearing as an input sample, taking radial and axial displacement as output samples, and collecting sample data;

selecting a mixed kernel function, optimizing performance parameters of a support vector machine through a particle swarm algorithm, training a least square support vector machine by using a training sample and the performance parameters, and establishing a nonlinear prediction model;

and step three, connecting the prediction model string with a linear closed-loop controller, and forming magnetic bearing displacement closed-loop control by combining an expanded current hysteresis three-phase power inverter and a switching power amplifier, so as to realize the self-detection of the three-degree-of-freedom alternating-current hybrid magnetic bearing displacement-free sensor.

The invention discloses a method for realizing displacement self-detection of a three-degree-of-freedom alternating current and direct current hybrid magnetic bearing by utilizing a hybrid kernel function support vector machine displacement prediction model.

Further, the first step specifically comprises:

continuous acquisition of q sets of current signals x in a magnetic bearing system with eddy current sensors_t＝[i_at,i_bt,i_ct,i_zt](t ═ 1,2,3.., q) and rotor displacement signal y_t＝[x_ft,y_ft,z_ft](t ═ 1,2,3.., q) as initial input and output sample data, where i is output as initial input_atIs the current signal of the magnetic pole A, i_btIs the current signal of the magnetic pole B, i_ctIs the current signal of the magnetic pole C, x_ftIs a displacement signal in the radial x-direction, y_ftIs a displacement signal in the radial y-direction, z_ftIs the axial z-direction displacement signal);

selecting q/2 groups as training sample set, current signal x_t＝[i_at,i_bt,i_ct,i_zt](t 1,2,3.., q/2) and a rotor displacement signal y_t＝[x_ft,y_ft,z_ft](t ═ 1,2,3.., q/2), for offline training samples; the other half q/2 group is used as a test sample set for testing the sample set.

Further, the second step is specifically as follows:

2.1, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt]Inputting the parameters into a particle swarm algorithm and a support vector machine respectively to determine parameters of a mixed kernel function in the support vector machine;

2.2, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt]Inputting the (t ═ 1,2,3.., q/2) into a particle swarm algorithm, optimizing a mixing coefficient lambda by adopting the particle swarm algorithm, reserving the lambda with the best performance index as an optimal performance parameter, and then inputting the lambda with the best performance index into a support vector machine;

2.3, q/2 sets of current signals x_t＝[i_at,i_bt,i_ct,i_zt]And (t 1,2,3, q/2) inputting into a support vector machine, and determining supportA mixed kernel function in a vector machine;

radial basis kernel function of K_l＝exp(-Px-x_tP²/2²) (t ═ 1,2,3.., q/2), where x is the input current signal, x is_tIs the input current signal of the training sample set, is the kernel width, by bringing the input current signal x into the test sample set_t＝[i_at,i_bt,i_ct,i_zt](t 1,2,3, q/2) obtaining a mapping characteristic map of the radial kernel function when the kernel widths respectively take different values, and constructing the radial kernel function K by using the one with the best reserved characteristic_l＝exp(-Px-x_tP2/2²)，(t＝1,2,3...,q/2)；

Polynomial kernel function of K_g＝((x,x_t)+1)^e(t ═ 1,2,3.., q/2), where x is the input current signal, x is_tIs the input current signal of the training sample set, e is the exponent, by bringing the input current signal x into the test sample set_t＝[i_at,i_bt,i_ct,i_zt](t ═ 1,2,3.., q/2.) can obtain a mapping characteristic map of the polynomial kernel function when the index e takes different values respectively, and the one e with the best retained characteristic constructs the polynomial kernel function K_g＝((x,x_t)+1)^e，(t＝1,2,3...,q/2)；

Forming a mixed kernel function K by the optimal radial basis kernel function and the optimal polynomial kernel function_m＝λK_l+(1-λ)K_g(t ═ 1,2,3.., q/2), wherein λ is a mixing coefficient with an optimal performance index;

2.4, q/2 current signals x_t＝[i_at,i_bt,i_ct,i_zt](t 1,2,3.., q/2) and a rotor displacement signal y_t＝[x_ft,y_ft,z_ft](t 1,2,3.., q/2) is substituted into the prediction output functionIn (2), the corresponding support vector coefficient a is obtained_t(t 1,2,3, q/2) and a threshold b, and further establishing a support vector machine reflecting the current-displacement relation of the magnetic bearingThe displacement prediction model can identify the prediction output of the output displacement according to the current input x of the model as

Wherein K_mAs a mixed kernel function, K_m＝λexp(-Px-x_tP²/2²)+(1-λ)((x,x_t)+1)^e(t ═ 1,2,3.., q/2), λ is a mixing coefficient.

Further, in the particle swarm optimization, each sample represents a particle, the size of the group is set as q/2, the q/2 th sample represents the q/2 th particle, and in a D-dimensional target search space, the coordinate position vector of each particle is expressed as

The velocity vector is expressed asThe individual extremum of the individual particles is recorded as

The optimal position searched by the particle is shown, and the global extreme value of the particle swarm is recorded asNamely the best position searched by the current particle swarm; the particle swarm optimization method comprises the following steps:

1) initializing a sample group and calculating a fitness function value;

2) in the implementation process of the particle swarm optimization, when a better current optimal solution which cannot be obtained by the particle swarm optimization is obtained later, a certain amount of constant disturbance is added, so that the inertia weight is suddenly increased in certain iteration, local search is skipped conveniently, global search is carried out, and local convergence is prevented; for the linear decreasing particle swarm algorithm for increasing the disturbance inertia weight, the particle swarm algorithm updating iterative calculation formula for introducing the inertia weight coefficient is as follows:

A_s∈{0,0.1}

wherein t 1,2, L, q/2 represents the number of the particles; s represents the s-dimension of the particle, s 1, 2.., D; d represents the number of iterations; c. C₁,c₂Taking a value between 0 and 2 as an acceleration constant; rand is a random real number of interval (0, 1); ω max is the initial maximum inertial weight; omega_kIs the decreasing slope of the inertial weight coefficient; a. the_sFor the inertial weight perturbation constant, at a perturbation probability of 10%, A_s0.1, the rest is A_s＝0；

3) The dynamic self-adaptive change inertia weight coefficient is as follows: under the probability of 40%, multiplying the inertia weight obtained by linearly decreasing the linear weight of the particle swarm algorithm for increasing the disturbance by a coefficient in a fixed range, wherein the coefficient r is in the interval of 0.9-1.1, namely r belongs to [0.9,1.1], and the specific dynamically adaptive particle swarm algorithm for changing the inertia weight coefficient updates an iterative calculation formula as follows:

A_s∈{0,0.1}

4) random individuals were introduced to maintain particle population diversity: according to the particle swarm updating mode, in the iterative process of the particle swarm algorithm, all individuals are close to the optimal particles, so that the particles of the particle swarm are aggregated to lose diversity; under the probability of 30%, correspondingly replacing a certain random individual in the solution space with the particles obtained by the particle swarm algorithm;

5) and obtaining the current optimal position, updating the particle position, judging whether the iteration stopping condition is met, if not, updating the particle position, recalculating the fitness function value, and if the iteration stopping condition is met, obtaining the optimal solution and outputting the optimal solution.

Further, the third step is specifically:

connecting a mixed kernel function support vector machine displacement prediction model with a corresponding linear closed-loop controller, and connecting a corresponding expanded current hysteresis three-phase power inverter and a switching power amplifier behind the linear closed-loop controller in series to jointly form displacement closed-loop control on the three-degree-of-freedom AC/DC mixed magnetic bearing, thereby realizing the self-detection of the three-degree-of-freedom AC/DC mixed magnetic bearing displacement-free sensor; current signal i of three-phase power inverter and switching power amplifier with extended current hysteresis by mixing kernel function support vector machine displacement prediction model_at，i_bt，i_ct，i_ztAs an input signal, a predicted rotor displacement x is output_ft，y_ft，z_ftAnd with a given reference position signal x_ft*，y_ft*，z_ftComparing, and outputting rotor suspension control force F by linear closed-loop controller_xt*，F_yt*，F_ztAnd the three-phase power inverter and the switching power amplifier are expanded to finally realize the stable suspension of the three-degree-of-freedom alternating current and direct current hybrid magnetic bearing.

The beneficial effects of the invention and the prior art are as follows:

1. because parameters in the particle swarm algorithm have great influence on the accuracy of the displacement prediction model of the magnetic bearing mixed kernel function support vector machine, the improvement of the particle swarm algorithm can be realized by optimizing the characteristics of the parameters in the particle swarm algorithm, and the particle swarm algorithm mainly accelerates the convergence speed in the initial stage of the population evolution based on the characteristics of the important parameters of the particle swarm algorithm; in the final stage of population evolution, the particle swarm algorithm mainly determines an accurate solution, wherein the inertial weight linear decreasing strategy is a mature optimization and improvement strategy of the particle swarm algorithm.

When the model parameters are optimized and trained, the particle swarm optimization algorithm is used and improved, a certain amount of constant disturbance is added on the basis of linear decrease of the inertia weight so as to jump out local search and perform global search, thereby preventing local convergence, and secondly, in order to keep the diversity of the particle swarm, random individuals are introduced with a certain probability, thereby avoiding the phenomenon of particle aggregation as much as possible, so as to improve the global convergence performance of the algorithm, obtain the optimal kernel function performance parameters, and further obtain a more accurate displacement prediction model of the magnetic bearing mixed kernel function support vector machine;

3. the method adopts a mixed kernel function support vector machine to establish a control current-radial axial displacement model of a controlled object, realizes the displacement-free sensor self-detection control of the three-degree-of-freedom alternating current-direct current mixed magnetic bearing system, reduces the axial size of a magnetic suspension bearing rotor, simplifies the system structure, improves the dynamic performance of the system, is particularly suitable for running in a high-speed occasion, omits some additional wires and physical instruments, and obviously reduces the overall cost price of the magnetic bearing;

4. the displacement prediction model is established by adopting a mixed kernel function support vector machine, the accurate mathematical model of the controlled system is not required to be known, and the magnetic bearing control system has a simple structure and is more beneficial to engineering practice. The support vector machine method adopts the structural risk minimization criterion on the basis of empirical risk minimization, better solves the problems of over-learning, dimension disaster, premature convergence and the like in the traditional machine learning methods such as a neural network and the like, and has better popularization performance.

Drawings

FIG. 1 is a schematic diagram of a model training of a three-degree-of-freedom AC/DC hybrid magnetic bearing support vector machine displacement prediction model;

FIG. 2 is a self-checking control block diagram of a three-degree-of-freedom AC/DC hybrid magnetic bearing;

FIG. 3 is a control block diagram of a three-degree-of-freedom AC/DC hybrid magnetic bearing system;

FIG. 4 is a particle swarm optimization flow chart of a three-degree-of-freedom alternating current-direct current hybrid magnetic bearing support vector machine displacement prediction model.

Detailed Description

In order to make the objects, technical solutions and effects of the present invention more clear, the present invention is further described in detail by the following examples. It should be noted that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The method takes the control current of the magnetic bearing as an input sample, radial and axial displacements as output samples, samples are collected, a mixed kernel function is selected, performance parameters of a support vector machine are optimized through a particle swarm algorithm, a least square support vector machine is trained by using a training sample and the performance parameters, a nonlinear prediction model is established, the prediction model is connected in series to the front of the three-degree-of-freedom alternating current and direct current mixed magnetic bearing and is connected with a linear closed-loop controller, an expanded current hysteresis three-phase power inverter and a switching power amplifier are added to form magnetic bearing displacement closed-loop control, and the self-detection of the three-degree-of-freedom alternating current mixed magnetic bearing.